For best results, apply to meta-llama/Llama-3-8B-Instruct with the "Midnight Enigma" parameter settings (text-generation-webui)....
Temperature: 0.98
top_p: 0.37
top_k: 100
typical_p: 1
min_p: 0
repetition_penalty: 1.18
frequency_penalty: 0
presence_penalty: 0
repetition_penalty_range: 1024
This model was trained using AutoTrain. For more information, please visit
AutoTrain.
1
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4model_path = "PATH_TO_THIS_REPO"
5
6tokenizer = AutoTokenizer.from_pretrained(model_path)
7model = AutoModelForCausalLM.from_pretrained(
8 model_path,
9 device_map="auto",
10 torch_dtype='auto'
11).eval()
12
13# Prompt content: "hi"
14messages = [
15 {"role": "user", "content": "hi"}
16]
17
18input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
19output_ids = model.generate(input_ids.to('cuda'))
20response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
21
22# Model response: "Hello! How can I assist you today?"
23print(response)
